Abstract: Uncertainty in each forecasting component, such as driving data, hydrological simulation, and early warning method, imposes an impact on the uncertainty of flash flood risk early warning, but its mechanism is not well understood yet and the theory and method for a quantitative combination analysis are lacking. This study develops a coupled uncertainty analysis method for flash flood early warning based on hydrological model simulations, formulates a dynamic critical rainfall index, and applies both to an analysis of the lower reaches of the Min River in Fujian Province. The results show the uncertainties caused by precipitation input, runoff simulation, and critical rainfall quantification account for 36%, 24% and 40% of the total uncertainty, respectively. Under a coupled uncertainty analysis, the average probabilities of flash flood disasters at risk levels IV, III, II and I are 58%, 65%, 79% and 81%, respectively. This new analysis method is of great significance in practical flash flood warming and disaster prevention.

Key words: flash flood warning, dynamic rainfall threshold, uncertainty analysis, Bayesian hierarchical modeling